Sunny Johari

Utilized LSTMs, RNNs, and advanced feature selection techniques like Boruta, SOMs, and XGB-DNN Fusion for fraud detection, identifying 85%+ critical anomalies, reducing manual investigation efforts by 60%. Implemented an LLM-powered SQL generation tool using Few-shot Prompting with RAG, enhancing query efficiency for anomaly detection. Led the design and implementation of a document comparison system using LLMs on Azure ML, streamlining workflows and accelerating decision-making. Derived business-critical insights using LLMs and operationalized them through dynamic Looker dashboards for executive decision-making. Led deployment and monitoring of fraud detection models using ONNX, ensuring scalability and performance tracking via PSI, CSI, and drift proxies. Developed merchant onboarding model using XGBoost and logistic regression, improving accuracy by 15% and optimizing retention strategies. Developed reusable modules for causal inference and discovery, integrating DAG learning, matching algorithms, and treatment effect estimation to streamline analytical workflows.

Developed an AI-driven personalized financial news recommendation system using NLP, clustering, TF-IDF ranking, and sentiment analysis to surface high-impact articles. Built a regression model for predicting advisor compensation based on AUM data. Packaged the code using Docker and deployed it on AWS EC2 for easy access and scalability. Designed state management system for FastAPI REST endpoint. Collaborated with managers and advisors to understand their needs and customize insights accordingly. Explored cutting-edge Generative AI use cases with Langchain and OpenAI to enhance data generation and creative solutions. Applied Named Entity Recognition (NER) and labeled the unstructured documents using AWS service such as Amazon Comprehend and AWS Ground Truth respectively. Built an abstractive summarization tool using Hugging Face Transformers library, generating summaries with a ROUGE score of over 0.5 and an F1-score of over 0.8.